Is PDSS a good way to analyze the Raptors defense?

[dagger]

[quote="chsh22"
I'm not sure of the following:
1) Why some people are somewhat quick to accept the measurements performed by a fan with admitted biases, but not those by a stats organization with questionable/unknown biases.
[/quote]

A stats organization? Are we talking MIT here? Cal Tech? If Boris is watching every game, every play, and applying his methodology, it shouldn't be dismissed because he doesn't have a web site and a catchy name and isn't trying to cover 30 teams. Individuals launch lots of good companies out of their basement.

[Clutch Carter]

BorisDK1, I don't know where you came from, but you are definitely a breath of fresh air.

[chsh22]

A stats organization? Are we talking MIT here? Cal Tech? If Boris is watching every game, every play, and applying his methodology, it shouldn't be dismissed because he doesn't have a web site and a catchy name and isn't trying to cover 30 teams. Individuals launch lots of good companies out of their basement.

Absolutely. By the same token anyone else's compiled stats shouldn't be thrown out the window unless you have sufficient cause to believe they are very flawed.

[Hendrix]

I do not think it's a good way to measure defense. on/off drtg is based around facts at least. It takes a significant sample size, and the results still need to be interpruted a bit, but atleast it gives you a good general idea without any manipulating of the stats.


PDSS's formula however is manipulated into producing a stat, and who knows exactly how things should be weighted in that stat to produce a good accurate result? The way it is though doesn't pass the smell test for me.

[dagger]

A stats organization? Are we talking MIT here? Cal Tech? If Boris is watching every game, every play, and applying his methodology, it shouldn't be dismissed because he doesn't have a web site and a catchy name and isn't trying to cover 30 teams. Individuals launch lots of good companies out of their basement.

Absolutely. By the same token anyone else's compiled stats shouldn't be thrown out the window unless you have sufficient cause to believe they are very flawed.

I think that one of the things this thread has shown is that there is more than one way to chart games, and the reason you have disputes between methodologies is that unlike baseball, there is more inter-action/inter-dependence/yin&yang between the players on the floor, hence gray areas occur in defensive coverage than no system can provide empirical answers to without a mindreader to understand intent.

Basketball is more subjective than baseball. And when methodologies clash, there has to be room for subjective OPINIONS to make up for the lack of absolutes. Its subjectivity vs objectivity. It's also the right to disagree and not be called crazy "because I know what I see".

[andreafan]

A stats organization? Are we talking MIT here? Cal Tech? If Boris is watching every game, every play, and applying his methodology, it shouldn't be dismissed because he doesn't have a web site and a catchy name and isn't trying to cover 30 teams. Individuals launch lots of good companies out of their basement.

Absolutely. By the same token anyone else's compiled stats shouldn't be thrown out the window unless you have sufficient cause to believe they are very flawed.

I think that one of the things this thread has shown is that there is more than one way to chart games, and the reason you have disputes between methodologies is that unlike baseball, there is more inter-action/inter-dependence/yin&yang between the players on the floor, hence gray areas occur in defensive coverage than no system can provide empirical answers to without a mindreader to understand intent.

Basketball is more subjective than baseball. And when methodologies clash, there has to be room for subjective OPINIONS to make up for the lack of absolutes. Its subjectivity vs objectivity. It's also the right to disagree and not be called crazy "because I know what I see".

True dag, in baseball stats don't generally lie, but in basketball stats with this complexity are still not without dispute.

[garbagnani]

entertaining debate.

Boris you have some very good points, you have done an excellent job of defending yourself against the ripp, reignman, supersub triple team lol.

seriously though. Boris' most valuable revelation is that we can't take stats at face, they don't always tell the whole story, especially in teamplay.

[andreafan]

entertaining debate.

Boris you have some very good points, you have done an excellent job of defending yourself against the ripp, reignman, supersub triple team lol.

seriously though. Boris' most valuable revelation is that we can't take stats at face, they don't always tell the whole story, especially in teamplay.

Two numbers matter in life, the day your born and the day you die.

[Crazy-Canuck]

Drtg tells me how poor the communication is between the defenders on our team.

PDSS tells me how poor individual defenders are overall and not on just one aspect of his defense.

Most of us here can agree that our guys looked confused out there.

Is it so far out of reach to say that our team is full of horrible defenders, but Bargs may be the 3rd or th best defender? We know he is a weak help defender, but he seems to be decent at everything else. That in it self should raise his overall rating.

[andreafan]

So this pdss stat puts andrea in good light, is this the essence of the argument ? Ha. Yup, tear down a stat that favors andrea's defensive prowess , par for the course i reckon.

[FirstInkTDot]

entertaining debate.

Boris you have some very good points, you have done an excellent job of defending yourself against the ripp, reignman, supersub triple team lol.

seriously though. Boris' most valuable revelation is that we can't take stats at face, they don't always tell the whole story, especially in teamplay.

Two numbers matter in life, the day your born and the day you die.

Your forgot the day you lose your virginity

[andreafan]

Without being born how can one go about losing one's virginity, that is the question.

[Ripp]

2. I don't see the noise diminishing, I see just a larger noisy sample being cobbled together. I'm not saying it's worthless - it does communicate something and we need to acknowledge that. But what does it communicate? That Bargnani himself is the ultimate cause of all badness, that there's no solution for Andrea? I don't think you can justify that stance, and certainly not from this kind of analysis.

You don't understand statistics, is the problem....one of the major themes of statistics is that with enough noisy samples, the underlying behavior of the generative process can be understood to exceedingly accurate accuracy by the appropriate algorithm/estimator/technique.

Let me illustrate with a small example. Suppose someone is flipping a coin. One side of the coin is marked 0, the other is marked 1. The coin has some probability "p" of coming up 1. Suppose he flips the coin N times. Assuming independence between coin flips, then you can estimate the number "p" to within accuracy on the order of 1/sqrt{N} (using the obvious estimator that just sets the estimate to the average number of 1s that were in the sequence of flips.)
So in other words, if he flips the coin 1000 times, your estimate of "p" is good to within a tolerance on the order of roughly .0316. If he flips it 1 million times, your estimate of "p" is good to within something on the order of 10^{-3}.

Look guys...when you use your cell phone and talk to your buddy, your voice gets converted to bits, and those are sent through the air. Well, the air is NOISY and random, and corrupts those bits sent! So if people (in this case, those good electrical engineers, statisticians and mathematicians who developed communication theory and information theory) did not know how to extract information from noisy information, lots of cool things you and I take for granted would not work.

So the noise argument is not going to fly, given that SS15 has 9000+ minutes worth of data. A bias argument might (bias is a bit more insidious, and I can illustrate this with a coin-flipping example too), but this certainly will not.

EDIT: And the independence assumption I made above can be dramatically weakened, so long as the correlation between coin flips isn't huge. So don't think it requires an assumption that doesn't show up in the real world.

[Ripp]

The flaw I see with on/off stats is that they don't take into account what the opposition is doing. It's not measuring the defense, it's measuring the opponent's offense. It's not a mirror, or a black and white thing.

But... it is a mirror, isn't it? If the other team scores a lot of points in the 4th quarter, then our 4th quarter defense is bad. If the other team starts scoring at an extremely high rate when our bench comes in, then our bench is bad defensively, no?

Yes, on/off sort of stats cannot explain why something is happening...e.g., why is the bench so terrible at defending, or why the starting lineup does X. But they can certainly can be used to tell you that something happened, that a lineup is good or bad. You first need to understand what you are having success or failure with first, before you can then diagnose why the failure is occurring, imo.

Like, we might find out that our bench is terrible defensively. So then we can look more deeply into this and find out why it is. Maybe because the guys on the bench are just inherently bad defenders. Or maybe they are average defenders, but just played a block of games against teams with 6MOY candidates, so our bench shouldn't get penalized for getting lit up by prime Manu or something. But we cannot start an investigation until we know who is doing well and who is doing poorly....then we can figure out the whys and the wherefores.

[Crazy-Canuck]

The flaw I see with on/off stats is that they don't take into account what the opposition is doing. It's not measuring the defense, it's measuring the opponent's offense. It's not a mirror, or a black and white thing.

But... it is a mirror, isn't it? If the other team scores a lot of points in the 4th quarter, then our 4th quarter defense is bad. If the other team starts scoring at an extremely high rate when our bench comes in, then our bench is bad defensively, no?

Yes, on/off sort of stats cannot explain why something is happening...e.g., why is the bench so terrible at defending, or why the starting lineup does X. But they can certainly can be used to tell you that something happened, that a lineup is good or bad. You first need to understand what you are having success or failure with first, before you can then diagnose why the failure is occurring, imo.

So how is that so different than PDSS?

Drtg can give us the numbers, but cant explain why? But everyone seems to just say Bargs is the root of the problem because thats what the numbers say. The numbers are right, but the interpretations are whats varied.

Who's to say our chemistry, communication or lack of schemes wasn't the problem? We cant know because these things arent countable.

What the PDSS tells me is that our entire team is bad up of bad defenders.

Both stats are helpful but doesnt answer any questions as to why our defense is bad, unless of course you believe that the onus is on the player and the player alone without any outside variables.

[Marlowe]

Hi Ripp,

Is it possible, through these analyses from Boris and supersub you have been able to quantify the concept of "poor chemistry". Here we have collection of individuals who seem to produce relatively positive individual stats yet collectively perform poorly relative to their individual capability? It's late I'm probably not thinking straight.

[Crazy-Canuck]

Hi Ripp,

Is it possible, through these analyses from Boris and supersub you have been able to quantify the concept of "poor chemistry". Here we have collection of individuals who seem to produce relatively positive individual stats yet collectively perform poorly relative to their individual capability? It's late I'm probably not thinking straight.

I agree that chemistry seemed to be a big issue and I think a lot of that has to do with Bargs having to anchor a defense for the first time in his basketball life, new faces and the lack of a leader/floor general.

Drtg of 5 man units should be a pretty good guage if you want to interpret them that way.

[Ripp]

The flaw I see with on/off stats is that they don't take into account what the opposition is doing. It's not measuring the defense, it's measuring the opponent's offense. It's not a mirror, or a black and white thing.

But... it is a mirror, isn't it? If the other team scores a lot of points in the 4th quarter, then our 4th quarter defense is bad. If the other team starts scoring at an extremely high rate when our bench comes in, then our bench is bad defensively, no?

Yes, on/off sort of stats cannot explain why something is happening...e.g., why is the bench so terrible at defending, or why the starting lineup does X. But they can certainly can be used to tell you that something happened, that a lineup is good or bad. You first need to understand what you are having success or failure with first, before you can then diagnose why the failure is occurring, imo.

So how is that so different than PDSS?

Drtg can give us the numbers, but cant explain why? But everyone seems to just say Bargs is the root of the problem because thats what the numbers say. The numbers are right, but the interpretations are whats varied.

Who's to say our chemistry, communication or lack of schemes wasn't the problem? We cant know because these things arent countable.

What the PDSS tells me is that our entire team is bad up of bad defenders.

Both stats are helpful but doesnt answer any questions as to why our defense is bad, unless of course you believe that the onus is on the player and the player alone without any outside variables.

Ah, but here is the difference. Suppose you calculate the number of points scored against the Raps in the 4th quarter of games, and compare it to the number of points scored in the first three quarters. You also count up the number of possessions used by opponents to get those points.
So you have the following table:

Code: Select all

Quarter Points Allowed Possessions Used
4th       1500                1000
Not 4th 3000                2500


So we eyeball this data and think, "Damn, why the hell is the other team scoring 1/3 of their points but using less than 30% of their possessions! Our defense really fell apart in the 4th!"

You then go ahead and divide, to normalize for possessions:

Code: Select all

Quarter Points Per 100 Possessions
4th       150
Not 4th 120


So this confirms your eyeballing of the data...the other team is scoring dramatically more efficiently in the 4th than in other quarters. Again, this is what actually happened, not some sort of statistic that is meant to estimate something. You divided total number of points scored by possessions used.

Now, somebody proposes some algorithm that will analyze players for you. It can also be used to analyze the performance of lineups, and thus of performance in different quarters. Suppose this algorithm is used, and it instead predicts the following 4th quarter performance, versus non-4th quarter performance:

Code: Select all

Quarter Points Per 100 Possessions
4th       80
Not 4th 160


In other words, it is saying that the team defense is better in the 4th than not in the 4th.

Will you still trust this algorithm that someone is trying to sell you? It is telling you that your team will be better defensively in the 4th than earlier, and this is in direct conflict with what actually happened, and you measured and saw in games. Personally, I would not trust such an algorithm...it would be evidence that there is something very wrong with the person's methodology.

Hopefully this illustrated what I mean. Looking at 4th quarter defense versus defense not in the 4th quarter is looking at what actually happened in games...the number of points scored against us, and how many possessions the opponent used to score those points. If you have an algorithm that produces results inconsistent with what actually occurred, no matter how beautiful, "right", and appealing the algorithm might be, then it has to be rejected out of hand.

[Ripp]

Hi Ripp,

Is it possible, through these analyses from Boris and supersub you have been able to quantify the concept of "poor chemistry". Here we have collection of individuals who seem to produce relatively positive individual stats yet collectively perform poorly relative to their individual capability? It's late I'm probably not thinking straight.

I don't think either of these approaches are able to quantify poor chemistry, or any other explanation for good or bad performance. What supersub15 has shown us is what actually occurred, not giving explanations for why it occurred (e.g., a player is a bad defender, some sort of bias against the player, poor chemistry between teammates, etc.)

That is something for us to think about and analyze. But like I said earlier, we cannot analyze things if we are working with faulty data, or drawing conclusions from a methodology that is inconsistent with reality (Boris's stat sheet.)

[Crazy-Canuck]

Cant both sets of data be right?

Your numbers tell us that the Raps were horrible as a unit last year. We are given those results without a causation.

Boris' numbers also tell us that our team was made up of bad individual defenders.

We can all agree that Jose was one of the worst defenders on our team last year. All the stats and eye ball test attest to this deduction.

The difference here is Bargs. Is he a bad defender overall as a player? Maybe Is a bad defender for the C position? Yes Does he do somethings well? Yes He is also replaced by the best defensive player on our team. So, I can see how Bargs would have rated decently individually as a player because there arent any stats that give added weight to the position.

Alot of people on this board cant make an unbiased opinion on anything regarding Bargs. If someone scores (unless its a straight iso) then there will be 2 people to blame regardless of who's right or wrong.

4th Q

I agree that it is easy to just use numbers to say he was a bad defender due to his on/off. But how can u definitively say its because he cant play defense or because the cohesion is bad?

I dont think Boris is saying the team will be better off without x or o player. His stats tell me that as an individual, x player doesnt fare as bad as people think. Now

In order to guage defense we cant just base it on players. We need to take into account of coaching, chemistry etc... But the problem is that we can measure any of those.